Your pump starts because the numbers add up
Sensors in the ground, a weather forecast that checks itself and a model of your field come together in Automations. That is where the decision is made: water, wait or do nothing.
Six links, from measurement to pulse
A timer waters because it is six o’clock. FairGrow waters because the soil, the weather and your field together say it is needed. This is how that works.
The sensor says how wet it is now
One soil moisture sensor per zone measures in the root zone. If the latest reading is too old, it does not count and nothing happens.
The weather, with an honest report card
FairGrow puts several weather models side by side and checks every forecast afterwards. If rain is on the way, the advice waits.
A model of your field
A water balance with your soil type, your slope and the Copernicus elevation model. Not how wet it is, but when it will get too dry.
The ML learns how your soil responds
For each pump and sensor, the model learns what a pulse does and where moisture settles. That way the control stops in time.
A rule you can read
On the canvas you draw lines between blocks, or you write a rule in YAML. Both are readable, and a new rule stays off until you switch it on.
The pump gets a pulse
Not a command that stays on, but short pulses that stop by themselves, within a daily cap and a rest period per zone.
A predictive model for your field
A sensor measures in one spot. Your field is bigger than that spot. So FairGrow works with the whole terrain: how water gets in, where it runs and which corner dries out first.
How much, how fast and where
The basis is a water balance following FAO-56: what comes in through rain and irrigation, what goes out through evaporation and how much fits in the root zone. Your own readings keep that balance honest.
On top of that comes the terrain. Your soil type and the slope inside your field boundary tell how many millimeters per hour the soil can take in. If the amount does not fit in one run, the advice says how many runs it does fit in. The Copernicus elevation model shows which corner stays wet longest.
That turns the advice into a forecast for your soil instead of an average for the region. Every threshold and assumption is on the irrigation advice page.
Learning how your soil responds to a pulse
After a run, the reading first shoots up and then drops back. Decide in that dip and you give too much water. The ML solves exactly that.
A model of its own for each pump and sensor
Water first runs quickly past the sensor and then slowly drains into the soil. The model learns those two speeds separately, for every combination of pump and sensor. With that, it predicts where moisture will end up once the soil has settled.
The model starts in shadow mode. It calculates along and records what it would have decided, but controls nothing. Only when it is demonstrably better than the raw reading does it take over. Otherwise the regular rule keeps doing the work.
If something in your setup changes, such as a new sensor position or a new pot, you reset learning for that one chain to zero. The old measurement history then no longer counts.
From canvas to a rule in YAML
For irrigation, you drag sensor, weather, decision engine and pump onto a canvas. Want to combine more? Write a rule in the same format as Home Assistant.
alias: Zone 1 water when the numbers add up# newly saved means off, until you switch it ontrigger: - platform: numeric_state entity_id: sensor.fg_zone1_soil_moisture below: 40condition: - condition: numeric_state entity_id: sensor.fg_wx_regen_24u below: 2action: - service: switch.turn_on target: entity_id: switch.fg_pump_zone1
Every measurement becomes usable
Your soil moisture, expected rain for the next 24 hours, VPD, growing degree days, leaf wetness and the satellite indices NDVI, NDMI and NDRE: in Automations, each is a value a rule can respond to. Per location too, if you have several sites.
Rules you already have in Home Assistant come along. FairGrow prepares a copy that is switched off, so you can first see what it does. Through the Home Assistant integration, FairGrow can also switch lights and switches that you release for it.
On the canvas, Check & simulate shows what a setup would do, and the YAML editor checks a rule before you save it.
Putting the forecast next to your own ground
FairGrow already checks every weather forecast afterwards. The next step is for your own sensors to become the referee, instead of a weather model for the whole area.
Your field as its own weather station
On Forecast verification you see how far ahead you can rely on the forecast: every forecast next to what actually happened, per source and per day ahead. For now, that reality comes from a reanalysis for the surrounding area.
A field in a hollow cools down more at night, a greenhouse behind a windbreak warms up faster. Your sensor measures that difference every day. We are building the step that checks and adjusts the forecast against it per field, so the model of your field also knows your own microclimate.
Part of this is already running: in the disease models, FairGrow adjusts the temperature and humidity forecast up to three days ahead with a sensor within 250 meters, once it has been measuring for a few days.
Better to miss a run than to flood a field
A system that keeps pumping because a sensor is broken does more damage than a system that does nothing. These locks sit in the core and no rule can switch them off.
The main switch is off
A zone starts by watching only. A valve opens only once you set it to live and turn on the main switch.
Pulses, never an open tap
If the connection drops at the wrong moment, the pulse stops by itself instead of running on.
An old reading is no reading
If the latest moisture reading is too old, nothing is pumped. Your screen tells you why it is quiet.
Stop when it rains
If it starts raining during a run, the run stops. If the forecast shows more than a few millimeters, the run is postponed.
Daily cap and rest period
Each zone has a daily maximum and a minimum time between two runs. A faulty sensor can never keep pumping.
Shadow mode first
New rules start switched off and the ML only calculates along at first. You see what would happen before anything happens.
What runs, what we are building and who we are looking for
The chain runs on a small scale on our own setup. It still has to prove itself on real fields with real growers.
The chain from sensor to pump
- Canvas with sensor, weather, decision engine and pump
- Water balance with soil type, slope and elevation model
- ML per pump and sensor, shadow mode first
- YAML rules and the Home Assistant integration
Your soil as the referee
- Checking the forecast per field against your own sensors
- Adjusting the forecast for the advice accordingly
- Letting the field model learn its own microclimate
Growers who watch along for a season
- A field or greenhouse with a pump or valve you can switch
- Willing to put the advice next to your own decision
- Feedback that decides what we build next
Short answers
Does FairGrow switch on my pump without me doing anything?
Only if you choose that. A zone is set to watching by default: FairGrow calculates, shows and records what it would do. Only once you set the zone to live and turn on the main switch does the system switch by itself, and then always in pulses within the daily cap.
What exactly does the ML learn?
How your soil responds to a pulse: how much water runs quickly past the sensor and how much stays. With that, the model predicts where moisture settles, so the control does not keep pumping in the dip between two runs. The model first calculates along in shadow mode and only takes over when it does better than the raw reading.
Why not just use a timer?
A timer does not know that 15 millimeters of rain fell last night, or that your sandy soil dries out faster than the clay down the road. FairGrow uses your reading, the weather and your terrain, and skips a run when it is not needed.
Which sensors do I need?
One soil moisture sensor per zone and a pump or valve you can switch, through Zigbee or through Home Assistant. The sensors page shows which sensor fits where.
Do you use my sensors to improve the weather forecast?
In the disease models, yes: there, a sensor within 250 meters adjusts the temperature and humidity forecast up to three days ahead. For the forecast itself and the irrigation advice, we are building that now.
Does this work with Home Assistant?
Yes. You bring existing automations along as a switched-off copy, and FairGrow can switch lights and switches that you release in the integration. More on bringing your Home Assistant automations.
Each link in depth
Do you have a field and a pump?
Watch along for a season. Set your zone to watching and put the advice next to what you would do yourself. Tell us about your field and we will think along about the setup, with no obligation.
